These are chat archives for ipython/ipython

12th
Apr 2015
Andreas Klostermann
@akloster
Apr 12 2015 10:10 UTC
@minrk, I used your requirements.txt, and I am now getting new errors when running ipython or jupyter
it seems like I am doing something horribly wrong ;-)
"ipython notebook" per se works, but "jupyter notebook" does not... I think I have to dig around in the docs some more....
Andreas Klostermann
@akloster
Apr 12 2015 11:04 UTC
also --profile-dir seems to have been discarded
Bas Nijholt
@basnijholt
Apr 12 2015 15:30 UTC
@minrk I start it via the http://localhost:8888/tree#clusters tap, and the client is supposed to be running on the hpc. The hpc works with PBS.
Min RK
@minrk
Apr 12 2015 18:34 UTC
@akloster there might be some kinks to work out in the installation from master, I'll see if I can get a full working dockerfile up and running soon. Jupyter will not inherit IPython's notion of a profile, so there will be no jupyter notebook --profile.
Damian Avila
@damianavila
Apr 12 2015 20:53 UTC
/@all, how we proceed with PR fixing things to be backported and forwardported. I probably has a fix for some issue in the slides, and it would be nice to have it in the 3.2 (or whatever next version) and also in the jupyter nbconvert new repo... Do I open two PRs duplicating the fix? I guess it is the best solution but maybe there is other way...
Anton Akhmerov
@akhmerov
Apr 12 2015 21:06 UTC
@jhamrick a nice post about deploying a jupyterhub, thanks for writing it up. Do you know how much would Rackspace charge for something of that scale? I couldn't make heads of tails of their pricing and offers.
Min RK
@minrk
Apr 12 2015 21:14 UTC
@damianavila yes, we aren't going to be able to do automated backports anymore. It will just have to be:
  1. make a PR against the new home
  2. after review, make a PR against IPython 3.x if it is agreed that is should be backported.
Damian Avila
@damianavila
Apr 12 2015 21:21 UTC
OK, I will do it in that way... thanks!
Min RK
@minrk
Apr 12 2015 21:22 UTC
@akhmerov @jhamrick has actually worked out the numbers, but I don't remember exactly. From here, using one OnMetal compute node for every 30 users, as Jess is doing, comes out to a bit under $20/student/month, or roughly $75/student/semester.
And for her class, that has an enormous amount of headroom.
I think with more price-conscious choices, she said it would be closer to $25/student/semester, depending on how many students and what resources they need.
Anton Akhmerov
@akhmerov
Apr 12 2015 21:33 UTC
@minrk thanks. And anything similar for tmpnb-type setup?
@minrk also on an unrelated note, is the absence of _repr_markdown_ a design decision, or was it just not high in the list?
Min RK
@minrk
Apr 12 2015 21:36 UTC
It's mostly the same, you just need to consider how much resources you want to give users.
Anton Akhmerov
@akhmerov
Apr 12 2015 21:36 UTC
You rent full-time servers?
Min RK
@minrk
Apr 12 2015 21:36 UTC
tmpnb.org, for instance, uses much less resources per user, and tmpnb only needs to consider how many concurrent users it will have.
and tmpnb easily fits on a single node, and the limiting facter is almost certainly memory, so it's pretty much a question of how much memory you want to give each ephemeral instance.
tmpnb.org runs closer to $1-5/user/month.
Anton Akhmerov
@akhmerov
Apr 12 2015 21:38 UTC
I see, thanks.
Min RK
@minrk
Apr 12 2015 21:38 UTC
yes, we have full-time servers.
Anton Akhmerov
@akhmerov
Apr 12 2015 21:42 UTC
What about the rendering markdown question? It just seems weird, that it's a part of existing infrastructure that's not really in API.
Min RK
@minrk
Apr 12 2015 21:42 UTC
markdown support was added much later than the rest
since the protocol supports arbitrary mime-types, the _repr_foo_ scheme isn't sufficient, so we wanted to define a new mechanism that would allow for arbitrary mime-types. Something like a _repr_mime_ mapping, but we still haven't gotten around to implementing that.
Anton Akhmerov
@akhmerov
Apr 12 2015 22:07 UTC
I see, makes sense. Thanks for the explanation.